Budgets
Model budgets
Set a daily spend cap per model and optionally pause a model entirely. Useful for limiting expensive models (e.g. claude-opus, gpt-4o) while leaving cheaper ones uncapped.
Setting a model budget
- Go to Budgets → Model budgets.
- Click Add model cap, select a provider and enter the model name.
- Enter a daily cap in USD. This is independent of the company daily budget.
- Click Save. The cap applies immediately.
Auto-pause
When a model's daily spend reaches its cap, Zelyx automatically pauses the model for the rest of the day. While paused, all calls to that model return HTTP 402 regardless of other budget state.
The model resumes automatically at midnight UTC when the daily cap resets. You can also resume it manually on the Budgets page before then.
Manual pause
You can pause a model at any time from the Budgets page, without setting a daily cap. This blocks all calls to that model immediately. Use this to:
- Temporarily disable a model during an incident
- Force agents to use a cheaper fallback while you investigate costs
- Block access to a model for compliance reasons
Model-level blocking on a Zelyx key
You can also restrict individual developers from using specific models by setting allowed models or blocked models on their Zelyx key (see API keys). This is a key-level policy, separate from the workspace-level model budgets.
The key-level check runs before the model budget check. If a key blocks gpt-4o, calls to that model are rejected with HTTP 403 (policy) rather than HTTP 402 (budget).